Research On Intelligent Management Of Autonomous Vehicle And Fleet Based On Data Analysis For Smart Cities

Smart cities(SCs)are an "ecosystem" that operates in a sustainable and intelligent manner.They rely on the coordinated work of intelligently connected devices and information and communication technologies(ICT)and use advanced data analysis technologies for monitoring and control,to optimize the utilized of limited resource.The large-scale deployment of telecommunications networks provides installation support for the construction of SCs;widely distributed sensors and wireless access devices collect a large amount of valuable data for the analysis of SCs;and continuous breakthroughs in machine learning(ML)technologies provided technical support for the big data processing,calculation analysis,and decision-making of SCs.An important part of SCs is the Intelligent Transport System(ITS)with autonomous driving fleet.Use big data analysis technology to mine the characteristics and patterns of urban traffic flow,study new theories and technologies for the rational and dynamic allocation of urban vehicle resources and explore new ways to solve the problems facing cities currently facing traffic congestion,energy consumption,and environmental pollution.Focusing on the needs of SCs,the paper studied the theory and technology of intelligent management of autonomous driving fleets based on data analysis.The main innovative research and results are as follows:1)This paper analyzes the situation of vehicular mobility models and gives a "demand-model-application" framework for Internet of Vehicles(IoV)/communication-transport network to solve practical problems of how to construct suitable vehicular mobility models.The paper also divides existing vehicular mobility models into vehicle distribution models,vehicle traffic flow models,and driving behavior models.It analyzes in detail the random pattern of vehicles in space,the traffic flow models corresponding to roadmaps,and individual driving behavior(such as with Cars and lane changes).Finally,modeling examples of different application scenarios including basic network connection analysis,offline network optimization,online network function implementation,and real-time autonomous driving are given.2)Aiming at the randomness of spatial-temporal prediction of user travel demand due to the influence of complex factors such as humanities and environment,this paper proposes a spatial-temporal prediction model of trip demand based on points of interest and a multi-variable long short-term memory algorithm.By analyzing the correlation between historical ride data and weather,regional facility distribution and other information,the influence of meteorological information on the reliability of historical travel data is considered in the time domain prediction,and the point of interest score is introduced in the spatial forecast to evaluate the regional facilities.Differences in regional ride demand caused by imbalances,and parameters of the model were trained using heterogeneous data sets.Compared with the univariate long short-term memory algorithm,the prediction accuracy of the multivariate long short-term memory model,that comprehensively considers meteorology,regional facilities,and user needs,is improved by 3.12%.Finally,this paper verifies the accuracy of the spatial-temporal prediction method using the Beijing taxi trajectory dataset.3)Aiming at the problem that users need to forecast the road conditions during planning paths to meet different needs of users,this paper provides a path planning algorithm based on regional capacity prediction.Build a heuristic pathfinding model(enhanced-A*)with both search efficiency and path accuracy based on application requirements.The cost function of the traditional A*algorithm is integrated with the regional speed estimation function,and the enhanced-A*algorithm optimizes the cost of avoiding congested sections in path planning based on the prediction of regional capacity.First,the back propagation neural network and the least square method were used to fit the actual traffic data of Beijing,and the linear relationship between the vehicular density and the average vehicle speed in the area was obtained.Secondly,for the enhanced-A*algorithm,a dynamically updated map of the additional area average vehicle speed feature is set to implement dynamic path planning.Simulation proves the effectiveness of the enhanced-A*algorithm for searching travel time optimized path and reduces the search time cost by about 25%.